Forward progress mechanism for stores in the presence of load contention in a system favoring loads

摘要

A multiprocessor data processing system includes a plurality of cache memories including a cache memory. In response to the cache memory detecting a storage-modifying operation specifying a same target address as that of a first read-type operation being processed by the cache memory, the cache memory provides a retry response to the storage-modifying operation. In response to completion of the read-type operation, the cache memory enters a referee mode. While in the referee mode, the cache memory temporarily dynamically increases priority of any storage-modifying operation targeting the target address in relation to any second read-type operation targeting the target address.

权利要求

What is claimed is:

1. A method in a multiprocessor data processing system including a cache memory of an affiliated processor core, the affiliated processor core, a plurality of remote processor cores, and an interconnect communicatively coupling the cache memory to receive operations initiated by the plurality of remote processor cores, the method comprising:

caching data in a data array of the cache memory, wherein the data includes a cache line associated with a target address;

maintaining a directory of the data held in the data array, wherein the directory includes coherence state information for the data;

the cache memory processing operations snooped from the interconnect by reference to the data array and the directory, wherein the processing includes:

in response to the cache memory snooping on the interconnect a first storage-modifying operation of the plurality of remote processor cores that specifies a same target address as that of a first read-type operation of the plurality of remote processor cores that was previously snooped from the interconnect is then being serviced by the cache memory, the cache memory determining whether the cache memory is designated by the coherence state information as responsible for servicing operations specifying the target address;

in response to a determination that the coherence state information designates the cache memory as responsible for servicing operations specifying the target address, the cache memory providing a retry response to the first storage-modifying operation;

in response to completion of servicing of the read-type operation by the cache memory, the cache memory entering a referee mode, and while in the referee mode, the cache memory maintaining designation of the cache memory by the coherence state information as responsible for servicing operations specifying the target address and temporarily dynamically increasing priority of any second storage-modifying operation of the plurality of remote processor cores snooped from the interconnect that specifies the target address in relation to any second read-type operation of the plurality of remote processor cores snooped from the interconnect that specifies the target address.

2. The method of claim 1, wherein:

temporarily dynamically increasing priority of any second storage-modifying operation comprises providing a retry response to the second read-type operation of the plurality of remote processor cores snooped from the interconnect that specifies the target address in absence a of the cache memory snooping any other operation specifying the target address.

3. The method of claim 1, further comprising:

the cache memory snooping, while in referee mode, the second storage-modifying operation specifying the same target address;

in response to the cache memory snooping the second storage-modifying operation specifying the same target address while in referee mode, the cache memory:

exiting referee mode, and

servicing the second storage-modifying operation at the cache memory.

4. The method of claim 3, wherein:

the first storage-modifying operation and the second storage-modifying operation are both initiated by a same remote processor core among the plurality of remote processor cores.

5. The method of claim 1, further comprising:

while in the referee mode, the cache memory detecting a timeout condition in absence of receiving any second storage-modifying operation specifying the same target address; and

in response to detecting the timeout condition, the cache memory exiting the referee mode.

6. The method of claim 1, wherein the cache memory only enters the referee mode in response to the coherence state information for the target address indicating a shared modified state.

说明书

This application is a continuation of U.S. patent application Ser. No. 13/368,958 entitled “FORWARD PROGRESS MECHANISM FOR STORES IN THE PRESENCE OF LOAD CONTENTION IN A SYSTEM FAVORING LOADS,” by Guy L. Guthrie et al., filed on Feb. 8, 2012, the disclosure of which is incorporated herein by reference in its entirety for all purposes.

BACKGROUND OF THE INVENTION

1. Technical Field

The present invention relates generally to data processing and, in particular, to servicing processor operations in a data processing system. Still more particularly, the present invention relates to dynamically increasing the priority of storage-modifying operations relative to read-type operations in data processing system.

2. Description of the Related Art

A conventional symmetric multiprocessor (SMP) computer system, such as a server computer system, includes multiple processing units all coupled to a system interconnect, which typically comprises one or more address, data and control buses. Coupled to the system interconnect is a system memory, which represents the lowest level of volatile memory in the multiprocessor computer system and which generally is accessible for read and write access by all processing units. In order to reduce access latency to instructions and data residing in the system memory, each processing unit is typically further supported by a respective multi-level cache hierarchy, the lower level(s) of which may be shared by one or more processor cores.

Cache memories are commonly utilized to temporarily buffer memory blocks that might be accessed by a processor in order to speed up processing by reducing access latency introduced by having to load needed data and instructions from system memory. In some multiprocessor (MP) systems, the cache hierarchy includes at least two levels. The level one (L1) or upper-level cache is usually a private cache associated with a particular processor core and cannot be accessed by other cores in an MP system. Typically, in response to a memory access instruction such as a load or store instruction, the processor core first accesses the directory of the upper-level cache. If the requested memory block is not found in the upper-level cache, the processor core then access lower-level caches (e.g., level two (L2) or level three (L3) caches) for the requested memory block. The lowest level cache (e.g., L3) is often shared among several processor cores.

Because multiple processor cores may request write access to a same cache line of data and because modified cache lines are not immediately synchronized with system memory, the cache hierarchies of multiprocessor computer systems typically implement a cache coherency protocol to ensure at least a minimum level of coherence among the various processor core's “views” of the contents of system memory. In particular, cache coherency requires, at a minimum, that after a processing unit accesses a copy of a memory block and subsequently accesses an updated copy of the memory block, the processing unit cannot again access the old copy of the memory block.

A cache coherency protocol typically defines a set of cache states stored in association with the cache lines of each cache hierarchy, as well as a set of coherency messages utilized to communicate the cache state information between cache hierarchies. In many cases, the cache states and state transitions of the coherence protocol are designed to favor read-type memory access operations over store-type operations. The prioritization of read-type operations over store-type operations can lead to forward progress issues for store-type operations in the presence of significant load contention.

SUMMARY OF THE INVENTION

A multiprocessor data processing system includes a plurality of cache memories including a cache memory. In response to the cache memory detecting a storage-modifying operation specifying a same target address as that of a first read-type operation being processed by the cache memory, the cache memory provides a retry response to the storage-modifying operation. In response to completion of the read-type operation, the cache memory enters a referee mode. While in the referee mode, the cache memory temporarily dynamically increases priority of any storage-modifying operation targeting the target address in relation to any second read-type operation targeting the target address.

BRIEF DESCRIPTION OF THE DRAWINGS

The novel features believed characteristic of the invention are set forth in the appended claims. The invention itself however, as well as a preferred mode of use, further objects and advantages thereof, will best be understood by reference to the following detailed description of an illustrative embodiment when read in conjunction with the accompanying drawings, wherein:

FIG. 1 is a high-level block diagram of an exemplary data processing system in accordance with one embodiment;

FIG. 2 is a more detailed block diagram of an exemplary processing unit in accordance with one embodiment;

FIG. 3 is a detailed block diagram of an L2 cache slice in accordance with one embodiment;

FIG. 4 is an exemplary timing diagram of the processing of a processor memory access operation in a lower level cache in accordance with one embodiment;

FIG. 5 is a high level logical flowchart of an exemplary process of servicing a processor memory access operation in accordance with one embodiment; and

FIG. 6 is a timing diagram of an exemplary process of servicing a processor memory access operation.

DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENT(S)

With reference now to the figures, wherein like reference numerals refer to like and corresponding parts throughout, and in particular with reference to FIG. 1, there is illustrated a high-level block diagram depicting an exemplary data processing system in accordance with one embodiment. The data processing system is depicted as a cache coherent symmetric multiprocessor (SMP) data processing system 100. As shown, data processing system 100 includes multiple processing nodes 102a, 102b for processing data and instructions. Processing nodes 102 are coupled to a system interconnect 110 for conveying address, data and control information. System interconnect 110 may be implemented, for example, as a bused interconnect, a switched interconnect or a hybrid interconnect.

In the depicted embodiment, each processing node 102 is realized as a multi-chip module (MCM) containing four processing units 104a-104d, each preferably realized as a respective integrated circuit. The processing units 104 within each processing node 102 are coupled for communication to each other and system interconnect 110 by a local interconnect 114, which, like system interconnect 110, may be implemented, for example, with one or more buses and/or switches.

As described below in greater detail with reference to FIG. 2, processing units 104 each include a memory controller 106 coupled to local interconnect 114 to provide an interface to a respective system memory 108. Data and instructions residing in system memories 108 can generally be accessed and modified by a processor core in any processing unit 104 of any processing node 102 within data processing system 100. In alternative embodiments, one or more memory controllers 106 (and system memories 108) can be coupled to system interconnect 110 rather than a local interconnect 114.

Those skilled in the art will appreciate that SMP data processing system 100 of FIG. 1 can include many additional non-illustrated components, such as interconnect bridges, non-volatile storage, ports for connection to networks or attached devices, etc. Because such additional components are not necessary for an understanding of the described embodiments, they are not illustrated in FIG. 1 or discussed further herein. It should also be understood, however, that the enhancements described herein are applicable to cache coherent data processing systems of diverse architectures and are in no way limited to the generalized data processing system architecture illustrated in FIG. 1.

Referring now to FIG. 2, there is depicted a more detailed block diagram of an exemplary processing unit 104 in accordance with one embodiment. In the depicted embodiment, each processing unit 104 is an integrated circuit including two processor cores 200a, 200b for processing instructions and data. In a preferred embodiment, each processor core 200 is capable of independently executing multiple hardware threads of execution simultaneously. As depicted, each processor core 200 includes one or more execution units, such as load-store unit (LSU) 202, for executing instructions. The instructions executed by LSU 202 include memory access instructions that request access to a memory block or cause the generation of a request for access to a memory block.

The operation of each processor core 200 is supported by a multi-level volatile memory hierarchy having at its lowest level a shared system memory 108 accessed via an integrated memory controller 106, and at its upper levels, one or more levels of cache memory, which in the illustrative embodiment include a store-through level one (L1) cache 226 within and private to each processor core 200, and a respective store-in level two (L2) cache 230 for each processor core 200a, 200b. In order to efficiently handle multiple concurrent memory access requests to cacheable addresses, each L2 cache 230 can be implemented with multiple L2 cache slices 230a1-230aN, each of which handles memory access requests for a respective set of real memory addresses.

Although the illustrated cache hierarchies includes only two levels of cache, those skilled in the art will appreciate that alternative embodiments may include additional levels (L3, L4, etc.) of on-chip or off-chip in-line or lookaside cache, which may be fully inclusive, partially inclusive, or non-inclusive of the contents the upper levels of cache.

Each processing unit 104 further includes an integrated and distributed fabric controller 216 responsible for controlling the flow of operations on local interconnect 114 and system interconnect 110 and for implementing the coherency communication required to implement the selected cache coherency protocol. Processing unit 104 further includes an integrated I/O (input/output) controller 214 supporting the attachment of one or more I/O devices (not depicted).

In operation, when a hardware thread under execution by a processor core 200 includes a memory access instruction requesting a specified memory access operation to be performed, LSU 202 executes the memory access instruction to determine the target real address of the memory access request. LSU 202 then transmits to hash logic 206 within its processor core 200 at least the memory access request, which includes at least a transaction type (ttype) and a target real address. Hash logic 206 hashes the target real address to identify the appropriate destination (e.g., L2 cache slice 230a1-230aN) and dispatches the request for servicing to the appropriate destination.

With reference now to FIG. 3, there is illustrated a more detailed block diagram of an exemplary embodiment of one of L2 cache slice 230a1-230aN (referred to generically as L2 cache slice 230a) in accordance with one embodiment. As shown in FIG. 3, L2 cache slice 230a includes a cache array 302 and a directory 308 of the contents of cache array 302. Although not explicitly illustrated, cache array 302 preferably is implemented with a single read port and single write port to reduce the die area required to implement cache array 302.

Assuming cache array 302 and directory 308 are set associative as is conventional, memory locations in system memories 108 are mapped to particular congruence classes within cache array 302 utilizing predetermined index bits within the system memory (real) addresses. The particular memory blocks stored within the cache lines of cache array 302 are recorded in cache directory 308, which contains one directory entry for each cache line. While not expressly depicted in FIG. 3, it will be understood by those skilled in the art that each directory entry in cache directory 308 includes various fields, for example, a tag field that identifies the real address of the memory block held in the corresponding cache line of cache array 302, a state field that indicate the coherency state of the cache line, a LRU (Least Recently Used) field indicating a replacement order for the cache line with respect to other cache lines in the same congruence class, and inclusivity bits indicating whether the memory block is held in the associated L1 cache 226.

L2 cache slice 230a includes multiple (e.g., 16) Read-Claim (RC) machines 312a-312n for independently and concurrently servicing load (LD) and store (ST) requests received from the affiliated processor core 200. In order to service remote memory access requests originating from processor cores 200 other than the affiliated processor core 200, L2 cache slice 230a also includes multiple snoop machines 311a-311m. Each snoop machine 311 can independently and concurrently handle a remote memory access request “snooped” from local interconnect 114. As will be appreciated, the servicing of memory access requests by RC machines 312 may require the replacement or invalidation of memory blocks within cache array 302. Accordingly, L2 cache slice 230a includes CO (castout) machines 310 that manage the removal and writeback of memory blocks from cache array 302.

L2 cache slice 230a further includes an arbiter 305 that controls multiplexers M1-M2 to order the processing of local memory access requests received from affiliated processor core 200 and remote requests snooped on local interconnect 114. Memory access requests, including local load and store operations and remote read and write operations, are forwarded in accordance with the arbitration policy implemented by arbiter 305 to dispatch logic, such as a dispatch pipeline 306, which processes each read/load and store request is processed with respect to directory 308 and cache array 302 over a given number of cycles.

L2 cache slice 230a also includes an RC queue 320 and a CPI (castout push intervention) queue 318 that respectively buffer data being inserted into and removed from the cache array 302. RC queue 320 includes a number of buffer entries that each individually correspond to a particular one of RC machines 312 such that each RC machine 312 that is dispatched retrieves data from only the designated buffer entry. Similarly, CPI queue 318 includes a number of buffer entries that each individually correspond to a particular one of the castout machines 310 and snoop machines 311, such that each CO machine 310 and each snooper 311 that is dispatched retrieves data from only the respective designated CPI buffer entry.

Each RC machine 312 also has assigned to it a respective one of multiple RC data (RCDAT) buffers 322 for buffering a memory block read from cache array 302 and/or received from local interconnect 114 via reload bus 323. The RCDAT buffer 322 assigned to each RC machine 312 is preferably constructed with connections and functionality corresponding to the memory access requests that may be serviced by the associated RC machine 312. RCDAT buffers 322 have an associated store data multiplexer M4 that selects data bytes from among its inputs for buffering in the RCDAT buffer 322 in response unillustrated select signals generated by arbiter 305.

In operation, processor store requests comprising a transaction type (ttype), target real address and store data are received from the affiliated processor core 200 within a store queue (STQ) 304. From STQ 304, the store data are transmitted to store data multiplexer M4 via data path 324, and the store type and target address are passed to multiplexer M1. Multiplexer M1 also receives as inputs processor load requests from processor core 200 and directory write requests from RC machines 312. In response to unillustrated select signals generated by arbiter 305, multiplexer M1 selects one of its input requests to forward to multiplexer M2, which additionally receives as an input a remote request received from local interconnect 114 via remote request path 326. Arbiter 305 schedules local and remote memory access requests for processing and, based upon the scheduling, generates a sequence of select signals 328. In response to select signals 328 generated by arbiter 305, multiplexer M2 selects either the local request received from multiplexer M1 or the remote request snooped from local interconnect 114 as the next memory access request to be processed.

Referring now to FIG. 4, there is depicted a time-space diagram of an exemplary operation on the interconnect fabric of data processing system 100 of FIG. 1. The operation begins with a request phase 450 in which a master 400, for example, an RC machine 312 of an L2 cache 230, issues a request 402 on the interconnect fabric. Request 402 preferably includes at least a transaction type indicating a type of desired access and a resource identifier (e.g., real address) indicating a resource to be accessed by the request. Common types of requests preferably include those set forth below in Table I.

TABLE I

Request

Description

READ

Requests a copy of the image of a memory block

for query purposes

RWITM (Read-

Requests a unique copy of the image of a memory block

With-Intent-To-

with the intent to update (modify) it and requires

Modify)

destruction of other copies, if any

DCLAIM (Data

Requests authority to promote an existing query-only

Claim)

copy of memory block to a unique

copy with the intent to update (modify) it

and requires destruction of other copies, if any

DCBZ (Data

Requests authority to create a new unique copy of a

Cache

memory block without regard to its present

Block Zero)

state and subsequently modify its contents;

requires destruction of other copies, if any

CASTOUT

Copies the image of a memory block from a higher level

of memory to a lower level of memory in preparation

for the destruction of the higher level copy

WRITE

Requests authority to create a new unique copy of a

memory block without regard to its present state

and immediately copy the image of

the memory block from a higher level memory to a

lower level memory in preparation for

the destruction of the higher level copy

PARTIAL

Requests authority to create a new unique copy of a

WRITE

partial memory block without regard to its

present state and immediately copy the

image of the partial memory block from a higher level

memory to a lower level memory in preparation

for the destruction of the higher level copy

Request 402 is received by snoopers 404, for example, snoopers 311a-311m of L2 cache slice 230a. In general, with some exceptions, snoopers 311 in the same L2 cache 230a as the master of request 402 do not snoop request 402 (i.e., there is generally no self-snooping) because a request 402 is transmitted on the interconnect fabric only if the request 402 cannot be serviced internally by a processing unit 104.

The operation continues with a partial response phase 455. During partial response phase 455, snoopers 404 that receive and process requests 402 each provide a respective partial response 406 representing the response of at least that snooper 404 to request 402. A snooper 404 within an integrated memory controller 106 determines the partial response 406 to provide based, for example, upon whether that snooper 404 is responsible for the request address and whether it has resources available to service the request. A snooper 404 of an L2 cache 230 may determine its partial response 406 based on, for example, the availability of its L2 cache directory 308, the availability of a snoop logic instance 311 within the snooper 404 to handle the request, and the coherency state associated with the request address in L2 cache directory 308.

The operation continues with a combined response phase 460. During combined response phase 460, the partial responses 406 of snoopers 404 are logically combined either in stages or all at once by one or more instances of response logic 422 to determine a system-wide combined response (referred to herein as “CR” or “Cresp”) 410 to request 402. In one preferred embodiment, which will be assumed hereinafter, the instance of response logic 422 responsible for generating combined response 410 is located in the processing unit 104 containing the master 400 that issued request 402, for example, in fabric controller 216. Response logic 422 provides combined response 410 to master 400 and snoopers 404 via the interconnect fabric to indicate the system-wide response (e.g., success, failure, retry, etc.) to request 402. If the CR 410 indicates success of request 402, CR 410 may indicate, for example, a data source for a requested memory block, a cache state in which the requested memory block is to be cached by master 400, and whether “cleanup” operations invalidating the requested memory block in one or more L2 caches 230 are required.

In response to receipt of combined response 410, one or more of master 400 and snoopers 404 typically perform one or more operations in order to service request 402. These operations may include supplying data to master 400, invalidating or otherwise updating the coherency state of data cached in one or more L2 caches 230, performing castout operations, writing back data to a system memory 108, etc. If required by request 402, a requested or target memory block may be transmitted to or from master 400 before or after the generation of combined response 410 by response logic 422.

In the following description, the partial response 406 of a snooper 404 to a request 402 and the operations performed by the snooper 404 in response to the request 402 and/or its combined response 410 will be described with reference to whether that snooper is a Highest Point of Coherency (HPC), a Lowest Point of Coherency (LPC), or neither with respect to the request address specified by the request. An LPC is defined herein as a memory device or I/O device that serves as the ultimate repository for a memory block. In the absence of a caching participant that holds a copy of the memory block, the LPC holds the only image of that memory block. In the absence of an HPC caching participant for the memory block, the LPC has the sole authority to grant or deny requests to modify the memory block. In addition, an LPC, when the LPC data is current and in the absence of a caching participant that can provide the data, provides that data to requests to either read or modify the memory block. If a caching participant has a more current copy of the data, but is unable to provide it to a request, the LPC does not provide stale data and the request is retried. For a typical request in the data processing system embodiment, the LPC will be the memory controller 106 for the system memory 108 holding the referenced memory block. An HPC is defined herein as a uniquely identified device that caches a true image of the memory block (which may or may not be consistent with the corresponding memory block at the LPC) and has the authority to grant or deny a request to modify the memory block. Descriptively, the HPC (even if its copy is consistent with main memory behind the LPC) also provides a copy of the memory block to a requestor in response to any request to read or modify the memory block (cache to cache transfers are faster than LPC to cache transfers). Thus, for a typical request in the data processing system embodiment, the HPC, if any, will be an L2 cache 230. Although other indicators may be utilized to designate an HPC for a memory block, a preferred embodiment designates the HPC, if any, for a memory block utilizing selected cache coherency state(s) within the L2 cache directory 308 of an L2 cache 230. In a preferred embodiment, the coherency states within the coherency protocol, in addition to providing (1) an indication of whether a cache is the HPC for a memory block, also indicate (2) whether the cached copy is unique (i.e., is the only cached copy system-wide), (3) whether and when the cache can provide a copy of the memory block to a master of a request for the memory block, and (4) whether the cached image of the memory block is consistent with the corresponding memory block at the LPC (system memory). These four attributes can be expressed, for example, in an exemplary variant of the well-known MESI (Modified, Exclusive, Shared, Invalid) protocol summarized below in Table II. Further information regarding the coherency protocol may be found, for example, in U.S. Pat. No. 7,389,388, which is hereby incorporated by reference.

TABLE II

Consistent

Cache

with

Legal concurrent

state

HPC?

Unique?

Data source?

LPC?

states

M

Yes

yes

Yes, before

no

I (& LPC)

CR

Me

Yes

yes

Yes, before

yes

I (& LPC)

CR

T, Te

Yes

unknown

Yes, after

no

S_L, S, I (& LPC)

CR if none

provided

before CR

S_L

No

unknown

Yes, before

unknown

T, S, I (& LPC)

CR

S

No

unknown

No

unknown

T, S_L, S, I (&

LPC)

I

no

n/a

No

n/a

M, Me, T, S_L, S,

I (& LPC)

Of note in Table II above are the T, S_L and S states, which are all “shared” coherency states in that a cache memory may contemporaneously hold a copy of a cache line held in any of these states by another cache memory. The T or Te state identifies an HPC cache memory that formerly held the associated cache line in one of the M or Me states, respectively, and sourced a query-only copy of the associated cache line to another cache memory. As an HPC, a cache memory holding a cache line in the T or Te coherence state has the authority to modify the cache line or to give such authority to another cache memory. A cache memory holding a cache line in the Tx state (e.g., T or Te) serves as the cache data source of last resort (after Cresp) for query-only copies of that cache line in that the cache memory will only source a query-only copy to another cache memory if no cache memory holding the cache line in the S_L state is available to serve as a data source (before Cresp).

The S_L state is formed at a cache memory in response to that cache memory receiving a query-only copy of a cache line from a cache memory in the T coherence state. Although the S_L state is not an HPC coherence state, a cache memory holding a cache line in the S_L state has the ability to source a query-only copy of that cache line to another cache memory and can do so prior to receipt of Cresp. In response to sourcing a query-only copy of a cache line to another cache memory (which assumes the S_L state), the cache memory sourcing the query-only copy of the cache line updates its coherency state for the cache line from S_L to S. Thus, implementation of the S_L coherence state can cause numerous query-only copies of frequently queried cache lines to be created throughout a multiprocessor data processing system, advantageously decreasing latencies of query-only access to those cache lines.

Referring again to FIG. 4, the HPC, if any, for a memory block referenced in a request 402, or in the absence of an HPC, the LPC of the memory block, preferably has the responsibility of protecting the transfer of ownership of a memory block, if necessary, in response to a request 402. In the exemplary scenario shown in FIG. 4, a snooper 404n at the HPC (or in the absence of an HPC, the LPC) for the memory block specified by the request address of request 402 protects the transfer of ownership of the requested memory block to master 400 during a protection window 412a that extends from the time that snooper 404n determines its partial response 406 until snooper 304n receives combined response 410 and during a subsequent window extension 412b extending a programmable time beyond receipt by snooper 404n of combined response 410. During protection window 412a and window extension 412b, snooper 404n protects the transfer of ownership by providing partial responses 406 to other requests specifying the same request address that prevent other masters from obtaining ownership (e.g., a retry partial response) until ownership has been successfully transferred to master 400. Master 400 likewise initiates a protection window 413 to protect its ownership of the memory block requested in request 402 following receipt of combined response 410.

Because snoopers 404 all have limited resources for handling the CPU and I/O requests described above, several different levels of partial responses and corresponding CRs are possible. For example, if a snooper within a memory controller 106 that is responsible for a requested memory block has a queue available to handle a request, the snooper may respond with a partial response indicating that it is able to serve as the LPC for the request. If, on the other hand, the snooper has no queue available to handle the request, the snooper may respond with a partial response indicating that it is the LPC for the memory block, but is unable to currently service the request. Similarly, a snooper 311 in an L2 cache 230 may require an available instance of snoop logic and access to L2 cache directory 406 in order to handle a request. Absence of access to either (or both) of these resources results in a partial response (and corresponding CR) signaling an inability to service the request due to absence of a required resource.

As discussed above, read-type operations are generally prioritized over store-type operations in data processing systems because the time critical path through a software program is generally determined by load latency. The prioritization of read-type operations over store-type operations can be expressed in a number of data processing system attributes, including the hardware architecture, memory model and coherency protocol implemented by a given data processing system. For example, the coherency protocol summarized in Table II reflects this prioritization by favoring the formation of numerous distributed query-only (S_L or S) copies of a frequently queried cache line throughout a multiprocessor data processing system. While the availability of numerous sources of a query-only copy of a cache line reduces access latency of non-storage-modifying operations, the presence of many copies of the cache line distributed throughout the data processing system can increase the access latency of storage-modifying accesses because any one of the cache memories holding a query-only copy of the cache line and servicing a request for query-only access can force a competing request for storage-modifying access to be retried. In certain cases, repeated retry of the storage-modifying access can slow or even halt forward progress of the program (e.g., if the storage-modifying access is required to release a highly contended lock).

In particular, in the coherence protocol described above, an HPC in a Tx state is required to protect the first read operation it sees and retry all other read and modification operations to that address until at least Cresp for that read. This is to ensure that in the absence of an S_L copy to source the read data (ether due to the S_L being busy with another request or being absent altogether), the HPC cache will provide the data. In the event that the HPC cache is modified relative to main memory in the LPC and the S_L cache is absent altogether, the HPC must perform this function as the LPC data is stale and the HPC cache must provide the data. In the other cases, it is preferential to have the HPC provide the data as the cache to cache transfer (even after Cresp), is more efficient than an LPC to cache transfer. As such, in the presence of significant numbers of querying read caches, a single cache attempting to store to a location (for example to release a lock) may have significant difficulty getting to the HPC and obtaining permission to modify the memory block since a significant portion of the time, the HPC is busy serving as the data source of last resort and is unable to lock on to the block modification request.

As described in detail below with reference to FIGS. 5-6, performance issues associated with read prioritization can be addressed by temporarily and dynamically increasing the priority of store-type operations relative to read-type operations so that a store may more effectively compete to be recognized by the HPC cache. In at least some embodiments, the priority of store-type operations are increased relative to read-type operations by forcing read-type operations targeting cache lines for which contention is detected to be retried.

Turning now to FIG. 5, there is illustrated a high level logical flowchart of an exemplary process by which a cache memory dynamically prioritizes the handling of store operations over read operations in accordance with one embodiment. For clarity, the flowchart of FIG. 5 will be described herein in conjunction with timing diagram 600 of FIG. 6.

The process depicted in FIG. 5 begins at block 500 in response to an L2 cache memory 230 that holds a target cache line of a read-type operation in a shared, modified coherence state (e.g., the Tx state) snooping the read operation from local interconnect 114. The read-type operation may be a READ operation as previously described or any other non-storage-modifying access to a target cache line. At block 502, dispatch logic 306 of the L2 cache memory 230 dispatches a snoop machine 311 to handle the read-type operation snooped from local interconnect 114, as depicted at reference numeral 602 in FIG. 6 by the transition of the snoop machine 311 from an idle state to a busy state. The dispatched snoop machine 311 accordingly begins protection for the target address of the snooped read-type operation. Said another way, while the read-type operation is being processed by the dispatched snoop machine 311, any subsequently received store-type operation while snooper is completing normal processing for the read, as depicted at reference number 606 in FIG. 6, having the same target address will receive a retry partial response from the L2 cache 230 (and consequently a retry Cresp that forces the master of the store-type operation to reissue the store-type operation in order to update the cache line associated with the target address). In one or more embodiments, the read-type operation and store-type operation have a target address that is associated with a shared modified cache state, for example, cache state “T” or “Te”, as described above.

As illustrated at block 504, if the snoop machine 311 has completed its processing of the read operation as depicted at reference numeral 604 of FIG. 6, the process proceeds to block 510, which is described below. If, at decision block 504, snoop machine 311 has not completed processing of the read operation, the process continues at block 505, which illustrates snoop machine 311 performing normal processing of the snooped read-type operation, which may include, for example, sourcing the target cache line to the requesting master by cache-to-cache intervention in response to receipt of a combined response indicating success of the read-type operation. At block 506, the snoop machine 311 determines whether or not it has detected a competing store-type operation, as depicted at reference numeral 606 of FIG. 6, that is, a store-type operation (i.e., any storage-modifying operation) that has the same target address as the snooped read-type operation and that was originated by another execution thread. As noted above, because the snoop machine 311 is already active processing a read-type operation directed to the target cache line, any conflicting store-type operation subsequently received while snoop machine 311 is processing the read-type operation will be given a retry partial response by the L2. In response to detecting a competing store-type operation at block 506, snoop machine 311 sets a flag to later enter a “referee” mode in which a store-type operation to the target cache line will be temporarily prioritized over any read-type operations to the target cache line, as depicted at block 508 of FIG. 5 and reference numeral 608 of FIG. 6. In one embodiment, the snoop machine only sets the flag to later enter the referee mode when the cache state associated with the target address is a shared modified state. Following block 508, the process returns to block 504, which has been described.

Referring now to block 510, after completing processing of the read-type operation, the snoop machine 311 determines whether or not the referee mode flag is set. If not, snoop machine 311 returns to an idle state, as illustrated in FIG. 5 at block 512 and in FIG. 6 at reference numeral 610. Following block 512, the process illustrated in FIG. 5 ends at block 513. Returning to block 510, if the snoop machine 311 determines the flag is set to enter referee mode, the snoop machine 311 does not return to the idle state upon completion of processing the read-type operation, but instead enters referee mode, as shown at block 514 of FIG. 5 and reference numeral 614 of FIG. 6. In conjunction with entering the referee mode, the snoop machine 311 also starts a timer, in one embodiment.

At block 516, the snoop machine 311 determines if it has detected a competing read-type operation, as depicted at reference numeral 618 of FIG. 6, directed to the same target cache line as that targeted by the competing store-type operation. If so, then the snoop machine 311 provides a retry partial response to the competing read-type operation, as shown at block 518 of FIG. 5. Thereafter, the process returns to block 516, which has been described.

If the snoop machine 311 has not detected a competing read-type operation at block 516, the snoop machine 311 determines at block 520 if a timeout of the referee mode has occurred. In various embodiments, a timeout may occur at a static predetermined timer value, or alternatively, at a dynamic value determined based on, for example, a number of competing read-type operations received during referee mode or when a ratio of a number of competing read-type operations to a number of store-type operations is reached. In response to determining at block 520 that the referee mode has timed out, the process proceeds to block 528, which illustrates the snoop machine 311 exiting referee mode (not depicted on FIG. 6). Thereafter, the snoop machine 311 is set to idle (block 512 of FIG. 5). If, however, the snoop machine 311 determines at block 520 that the referee mode has not timed out, snoop machine 311 further determines at block 522 whether a store-type operation targeting the target address of the previously processed read-type operation has been received. In the case of significant read contention against a single store releasing a lock, any such store-type operation most likely represents the master of the previous store-type operation that received a retry partial response reissuing a store-type operation to the same target address. If snoop machine 311 detects a store-type operation to the target address of the previously processed read-type operation, the process proceeds to block 524, described below. If, however, the snoop machine 311 does not detect a store-type operation to the same target address as that of the previously processed read-type operation, the process returns to block 516, which has been described.

Referring now to block 524, the snoop machine 311 exits referee mode, meaning that read-type operations are again prioritized above store-type operations. The exit from referee mode is illustrated at reference numeral 612 of FIG. 6.

At block 526, the snoop machine 311 processes the store-type operation normally, for example, by invalidating the target cache line in directory 308, and depending on the type of store-type operation, providing a copy of the target cache line to the requesting master in response to receipt of a Cresp indicating success of the store-type operation. Processing of the store-type operation by a different snoop machine 311 is depicted in FIG. 6 at reference numeral 630. Those skilled in the art will realize that in other embodiments, it would be possible to have the same snoop machine that was in referee mode also process the store 626 of FIG. 6. Whether it is feasible or advantageous to do so depends on specific implementation issues that vary from embodiment to embodiment. Thereafter, the process passes to block 512, which has been described.

In alternative embodiments of the process of FIG. 5 having more granular prioritization between read-type and store-type operations, snoop machine 311 may elect to force the retry of less than all of the competing read-type operations received while in referee mode (e.g., one-half or three-fourths). Further, in at least some embodiments, the scope of retry enforced by the referee mode could extend to a memory region larger than a single cache line (e.g., a memory page or other sized memory block). Further, in at least some embodiments, the duration of the referee mode may extend beyond receipt of a single store-type operation for the target cache line for which contention exists.

As has been described, a multiprocessor data processing system includes a plurality of cache memories including a cache memory. In response to the cache memory detecting a storage-modifying operation specifying a same target address as that of a first read-type operation being processed by the cache memory, the cache memory provides a retry response to the storage-modifying operation. In response to completion of the read-type operation, the cache memory entering a referee mode. While in the referee mode, the cache memory temporarily dynamically increases priority of any storage-modifying operation targeting the target address in relation to any second read-type operation targeting the target address.

While various embodiments have been particularly shown and described, it will be understood by those skilled in the art that various changes in form and detail may be made therein without departing from the spirit and scope of the appended claims and these alternate implementations all fall within the scope of the appended claims.